Data Center Capacity Is Not One Number

Data center capacity is one of the most repeated numbers in our market and one of the least comparable. A project is described as 40 MW, 120 MW or 640 MW, and the figure quickly becomes shorthand for scale, value and readiness. Yet a data center capacity statement can refer to installed IT load, utility connection capacity, a long-term campus design, or merely the first phase a developer intends to build.

That distinction is not semantic housekeeping. It changes how an investor values land, how an operator plans deployment, how a lender tests completion risk and how an end user judges whether capacity will actually be available when needed. In PowerlandMap market intelligence, we therefore treat capacity as an evidence chain, not a single field.

Start with four different megawatts

The cleanest way to read a capacity claim is to ask what physical boundary the number describes.

Installed IT load is the power that can be delivered to computing equipment. It is usually the most commercially useful denominator because customers contract racks, halls or suites around IT load. The European Commission's data-centre reporting regulation of 14 March 2024 also uses installed information-technology power as a core classification measure. That does not make every published MW figure an IT figure; it shows why the basis must be named.

Utility or grid capacity sits upstream. It may be expressed in MW, MVA or a contracted connection envelope. Conversion is not automatic: power factor matters, and not every unit of incoming electrical capacity reaches the IT equipment after redundancy, transformation, cooling and auxiliary loads.

Campus design capacity is the maximum scale contemplated by the master plan. It can be valuable evidence of ambition and land-use potential, but it may span several buildings, connection stages and investment decisions. It should not be treated as today's available capacity.

Deliverable phase capacity is what a specific building or phase can plausibly energise by a defined date. For a buyer, this is often the decisive number. Our development-readiness framework focuses on the milestones behind that date rather than the headline alone.

These four numbers can all be accurate. The problem begins when they are placed in the same comparison table without labels.

Recent announcements show why the basis matters

Consider three public examples.

Princeton Digital Group said in its 24 April 2025 TY1 launch announcement that the Saitama campus has 96 MW of IT capacity. The basis is explicit: IT capacity. That gives an analyst a comparatively clean starting point, although building-level availability and contracted load still require separate evidence.

Vantage described its Osaka KIX1 campus in a 22 May 2024 announcement as offering up to 68 MW, while stating that the first facility would include 28 MW of IT load. One campus, two useful numbers, two different scopes. Collapsing them into “68 MW available” would erase the phasing that matters commercially.

The contrast is sharper in West Java. BDx announced on 22 September 2026 a 640 MW campus backed by 845 MVA of secured grid power, while the first building is stated at 120 MW of IT capacity to be delivered in phases. Those figures describe campus scale, upstream power and an initial building. They are complementary, not interchangeable.

A fourth example adds contractual maturity. Brightray's 25 September 2026 Macao announcement describes a design for approximately 40 MW of IT capacity, but the framework does not itself oblige either party to buy or supply modules and depends on tenant and lease conditions. The number is an engineered target, not proof of a committed delivery.

Fact: these announcements state different capacity bases and different stages.

Inference: the economic value of each megawatt depends on the evidence supporting land, connection, design, construction, contracting and timing.

My view: capacity should be recorded as a vector — basis, amount, phase, date and confidence — before it is ever used as a ranking metric.

How to underwrite data center capacity

I use a simple evidence ladder. It is deliberately stricter than a marketing database.

At the first rung sits a conceptual or announced campus figure. It is useful for mapping future supply, especially when combined with our global coverage, but it carries little delivery certainty on its own.

Next comes evidence of site control and planning. A credible land position, permitted use and a design that fits the parcel move the project beyond a press-release polygon. The questions overlap with powered-land due diligence: who controls the land, what is actually permitted, and which constraints remain unresolved?

The third rung is power evidence. A substation nearby is not the same as a connection offer. A utility letter is not necessarily a fully executed agreement. A reservation is not always deliverable on the desired date. Costs also matter; our analysis of grid-upgrade cost allocation explains why a technically possible connection can still be commercially weak.

The fourth rung is financed and procured delivery. Contractors, long-lead equipment, civil works and commissioning plans make a phase more tangible. This is where commissioning readiness becomes relevant: energisation is a milestone, not the same thing as tested customer-ready capacity.

The final rung is commercially committed capacity with a credible counterparty and binding documents. Even then, analysts should separate a signed pre-lease from an operating hall. Our recent note on data center pre-lease agreements sets out the conditions that still need to be tested.

The practical test is a timeline

When I review a capacity claim, I want five answers in one line:

1. What is the unit and capacity basis? 2. Which building or phase does it cover? 3. What public document supports it? 4. What has to happen before it becomes deliverable? 5. On what date could a customer actually use it?

If one answer is missing, the number should not disappear. It should carry a confidence label. This is especially important in less transparent markets, where reported campus capacity can outpace publicly evidenced delivery. The same discipline underpins our work on African data center capacity evidence.

It also matters for capital. Large financing packages can support a platform rather than a named phase, while leases, guarantees and residual-value structures may sit at different entities. Capacity should therefore be linked to the relevant asset and obligation, not inferred from corporate funding alone. The distinction is central to our analysis of AI data center financing.

The power system will make this discipline more important, not less. The IEA reported on 16 April 2026 that data-centre electricity use rose 17% in 2025 amid tightening bottlenecks. As grid queues lengthen, two projects with the same announced MW can have radically different delivery paths.

Data center capacity is an evidence problem

A megawatt remains a precise engineering unit. A data center capacity claim is not precise until its boundary, maturity and date are clear.

For developers, that means publishing the basis rather than relying on scale alone. For investors and lenders, it means underwriting phases instead of multiplying a campus headline by a market multiple. For operators and end users, it means comparing deliverable IT capacity on the required date, with the power and commissioning evidence attached.

That is the practical purpose of a rigorous data center capacity model: not to make every announcement look smaller, but to make different projects genuinely comparable. PowerlandMap connects those claims to sites, power, phases and sources. Teams that need to test a market or shortlist can request access to review the evidence behind the map.

*Matthieu Gallego*